AI Briefing#karpenter
The latest AI and developer news about #karpenter, with the original source and a short summary.
The latest AI and developer news about #karpenter, with the original source and a short summary.
Ablely managed the lifecycle of AI agent workloads using Amazon EKS and Agent Sandbox.
Netflix increased job success rates from ~98% to above 99.5% by implementing layered health checks and automatic remediation for GPU infrastructure.
The Yeogi Eottae SRE team combined AI and code to automate infrastructure cost estimation and EKS node scaling predictions.
DelightRoom migrated 5 clusters to EKS Auto Mode, significantly reducing operational burden.
Automated Agentic AI operations using Amazon EKS Auto Mode, Bifrost, and Langfuse.
Using PodAffinity and do-not-disrupt, we auto-scaled EKS node groups for Job workloads.
By moving from SQS+Lambda to an EKS Spot-based setup, they achieved 70-90% cost savings and improved processing speed.
A missing libdcgm.so on Bottlerocket GPU nodes was causing Karpenter to repeatedly replace them.
Using Bottlerocket data volume snapshots, image pull time was reduced from 15 minutes to around 10 seconds.
By separating Karpenter from the ASG and applying constraints and reservation settings, they achieved both stability and cost savings.
Kurly built its own MLOps platform using Karpenter and NVDP to implement automatic allocation and deprovisioning of GPU resources.